Docs
CommunityJev Benchmarks & Evaluations
Bekko System One Dataset v0
hotchpotch's training dataset for the Bekko System One v0 decision family: a 1M–10M-row parquet collection whose configs are named for the public datasets behind it, tagged typed-decisions, choice, noul and score.
- Format
- Docs
- Reading time
- 2 min
- Published by
- Community
- Author
- hotchpotch
- Last verified
- Oct 2026
Highlights
- Size category 1M–10M rows, parquet, text modality; repository tags include typed-decisions, choice, noul and score.
- Config names map to public corpora: arc, boolq, bbq, banking77, clinc, civil_comments, aqua_rat and many more.
- Retrieval-style configs such as bge-hn-msmarco, bge-hn-squad and bge-hn-pubmed_qa_labeled ship as their own splits.
- Listed with 5 likes and 1,477 downloads; created 2026-09-28 by the same author as the Bekko System One v0 models.
- The digest's card extract is the config listing, cut off after 40-plus configs; no prose description appears in it.
Watch out
Card reviewed 2026-10-17; the digest's extract of this card is truncated to the config listing, so purpose, licence and provenance are unverified — no licence is stated and each source dataset keeps its own terms. Claims here are limited to the visible config names and metadata.
Questions about this entry
Short answers drawn from the linked project or page.
hotchpotch's dataset repository for Bekko System One v0 training data: a large parquet collection whose config names match public datasets such as arc, boolq, banking77, clinc and civil_comments.
The tags are typed-decisions, choice, noul and score; choice, noul and score are the decision kinds used across the Bekko System One v0 model family's contract.
The listed size category is 1M–10M rows in parquet with text modality, but the digest extract does not show the split counts, and the repository metadata lists 1,477 downloads at the time of review.
Configs such as bge-hn-msmarco, bge-hn-squad, bge-hn-nq and bge-hn-trivia hold retrieval-style training pairs, sitting alongside classification, NLI and benchmark-style corpora like arc, boolq and banking77.
No licence tag or statement appears in the extract, so reuse terms are unresolved; treat the material as carrying each source dataset's own terms until the repository card is checked directly.
It is published by the same author as the Bekko System One v0 17M, 68M and 400M checkpoints; the digest provides the config listing only, so the training recipe is not documented here.
Purpose, provenance, licence and split sizes: the extract is a truncated config listing, so stronger claims about composition or intended use would go beyond what the card as delivered shows.
More Jev resources like this
The research record of the Xyntetik typed-decisions track: TD-000 measurements and the preregistered TD-001 and TD-002 tests of whether /v1/decide probabilities from Xyntetik-Kvist-14B are usable as decisions. Both tests are INCONCLUSIVE; the record holds questions, scored rows, analyses and the harness. Apache-2.0.
Jev Benchmarks & EvaluationsDocs#community#evaluation#calibration
Community
xiq's JEV-format view of OpenJev-Data: every browser step re-expressed as state, questions and gold across eight configs covering Mind2Web, NNetNav, WebChain, two teacher runs and on-policy GLM, plus a frozen 2,346-step bench and 153 live tasks. Licences are mixed and per-row terms apply.
Jev Benchmarks & EvaluationsDocs#community#evaluation#browser-agent
Community
johannhartmann's German translation of the LocalLLaMA/typed-decisions benchmark at revision c806f6c: states, instructions, options and rubrics translated to German while keys, ids and gold distributions stay as in the original. Machine-translated without human review. Apache-2.0.
Jev Benchmarks & EvaluationsDocs#community#evaluation#typed-decisions
Community
Twelve frozen typed-decision configs from the independent Open-Jev project, publishing controlled and mostly synthetic tasks with probability or ordinal targets: from the 113,568-row default release to the 160,797-row browser and drone expansion. CC0 1.0; English, Chinese and Turkish.
Jev Benchmarks & EvaluationsDocs#community#evaluation#typed-decisions
Community
326,619 typed decision records from the frozen community-hard-mix-v2-final mixture: a redistributable projection of the data behind Open-Jev 27B training, with 2,053 Wikispeedia rows omitted.
Jev Benchmarks & EvaluationsDocs#community#benchmark#evaluation
Community
LangWatch Instant Evals ships a TypeSafe Jev classifier that batches Noul, Score, and Choice judgments on production traces in one POST per text.
Jev Benchmarks & Evaluations#community#typescript#evaluation
Community
langwatch/langwatchTypeScript
langwatch/langwatchTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.